Gravel aggregate grading rapid detection method based on image recognition

Abstract

The invention discloses a gravel aggregate grading rapid detection method based on image recognition, which adopts an on-line detection method and comprises the following steps: S1, image form extraction, including the following sub-steps: S1.1, aggregate cleaning; the method comprises the following steps: S1.2, acquiring an image which contains a reference object, and performing data normalization processing and view angle transformation by using space coordinates to generate a high-robustness gravel aggregate feature map; the method comprises the following steps: S1.3, image preprocessing: denoising, and then preprocessing the image by adopting a histogram equalization method; the method comprises the following steps: S1.4, carrying out image binaryzation; step S1.5, determining the boundary of the image, and extracting the particle size distribution of the gravel aggregate; and S2, grading detection: analyzing the form of the aggregate through image recognition. According to the method, morphological analysis is carried out on the gravel aggregate through image recognition, and the gradation characteristics of the gravel aggregate of the vibro-replacement gravel pile can be rapidly and accurately obtained in the construction process.

Description

Technical Field

[0001] This invention relates to a rapid detection method for sand and gravel aggregate gradation, and in particular to a rapid detection method for sand and gravel aggregate gradation based on image recognition. Background Technology

[0002] Gradation is a key factor affecting the mechanical properties of compacted aggregate in vibro-compacted stone piles. In traditional construction management, gradation strips for single-pile tests are pre-determined based on different pile formation methods during the initial design phase, and construction material distribution is carried out accordingly. After the stone piles are constructed, the aggregate is extracted from the pile body through excavation, and gradation curves are obtained through sieve analysis. Although manual sieving or shaking sieve technology is mature, this traditional method is cumbersome, time-consuming, and lacks representativeness. Especially in the testing of large quantities of aggregate, manual methods are inefficient, easily affected by human factors, and the test results lack representativeness. More seriously, this inspection is usually a post-construction inspection; if the aggregate gradation is found to be non-compliant, reconstruction is required, wasting significant time, money, and manpower. Therefore, how to quickly and accurately obtain the gradation characteristics of aggregates during construction has always been a key concern in the engineering field.

[0003] Image recognition technology offers a potential solution to this problem. While classic image processing algorithms can identify particle size in some cases, the complex morphology and uneven texture of sand and gravel aggregates often lead to oversegmentation or undersegmentation, limiting the accuracy of the recognition results. Therefore, improving the accuracy of image recognition technology, especially in complex field environments, remains a pressing issue.

[0004] Although image recognition technology has been widely used in many engineering fields, research on the rapid detection of aggregate gradation in vibro-compacted stone piles is still relatively scarce. The main reasons are as follows: (1) The particle size of sand and gravel aggregate varies greatly and the shapes are different, which makes the detection difficult. (2) The captured images may contain impurities, overlaps, noise or artifacts, and uneven lighting may cause changes in image brightness; (3) If the focal length or scale is not fixed during the shooting process and the image lacks obvious boundaries, the accuracy of particle size analysis is insufficient.

[0005] These external factors reduce the contrast between the target and the background, significantly affecting the effective analysis of sand and gravel aggregate images. Therefore, rapid detection of sand and gravel aggregate gradation in vibro-compacted stone piles remains a challenging task. Summary of the Invention

[0006] This invention provides a rapid detection method for sand and gravel aggregate gradation based on image recognition to solve the technical problems existing in the prior art. It can quickly and accurately obtain the gradation characteristics of sand and gravel aggregate in vibratory compaction stone piles during construction.

[0007] The technical solution adopted by this invention to solve the technical problems existing in the prior art is: a rapid detection method for sand and gravel aggregate gradation based on image recognition. During the construction of vibratory compaction stone piles, a high-pressure spray device, an industrial camera located downstream of the high-pressure spray device, a reference object of known size, and supplementary lighting equipment are fixedly installed on the sand and gravel aggregate conveying equipment. Then, the following steps are used for detection: Step S1, Image Morphology Extraction, includes the following sub-steps: Step S1.1: Aggregate cleaning. Turn on the high-pressure spray device to rinse the surface of the aggregate with water. Step S1.2: Image acquisition. An industrial camera captures images of sand and gravel aggregates. The images contain reference objects to determine the scale. Using spatial coordinates, the angles of each aggregate node to the origin are calculated. Data normalization and perspective transformation are performed to generate a highly robust sand and gravel aggregate feature map. Step S1.3: Image preprocessing, first remove noise, and then use histogram equalization to preprocess the image. Step S1.4: Image binarization. The K-means algorithm is applied to cluster the sand and gravel aggregate image, dividing the pixels into two categories: target and background. The target is the sand and gravel aggregate. The gray value of the cluster center is used for image binarization. Step S1.5: Image boundary determination. The eight-chain code algorithm is used to find connected regions in the image, fill the holes of the target, remove noise inside the target, mark the boundary of the target, and extract the particle size distribution of sand and gravel aggregate. Step S2, gradation detection: The aggregate morphology is analyzed by image recognition, the equivalent diameter of the fitted ellipsoidal aggregate particles is calculated, and the volume and mass of the particles are calculated by combining the scale and geometric features. Finally, the gradation curve of the vibratory compaction stone pile sand and gravel aggregate is drawn.

[0008] Based on the above solution, the present invention has made the following improvements: The industrial camera includes multiple industrial cameras with different shooting angles. In step S1.2, the multiple industrial cameras take pictures of sand and gravel aggregates from different angles, perform spatiotemporal alignment processing on the obtained image data, and then obtain a spatial distribution map of the aggregates.

[0009] In step S1.2, the spatial distribution map of the aggregate is obtained using an image, and the spatial location of the aggregate is expressed using the following formula based on a spatial coordinate system:

[0010] In the formula: (x,y,z) represents the three-dimensional spatial coordinates; n represents the number of nodes; O represents the origin of the three-dimensional space; By calculating the lines connecting each node of the aggregate to the origin. S nO With the direction of the conveyor belt movement S OV The angle between F ( n , t We obtain the angle formula, which is expressed as follows:

[0011] The L(n,t) and F(n,t) data are normalized and transformed by perspective. After angle adjustment and perspective transformation, a highly robust sand and gravel aggregate feature map is generated.

[0012] In step S1.3, the histogram equalization process uses the cv2.equalizeHist() function to map pixel values ​​using a cumulative distribution function, the expression of which is:

[0013] In the formula: MN is the total number of pixels in the image; CDFmin is the minimum value of the cumulative distribution; L is the number of gray levels.

[0014] The calculation formula for the K-means algorithm in step S1.4 is as follows:

[0015] In the formula: v i For the first i Cluster centers of a class; | x j - v i |for the first j From the pixel to the i Euclidean clustering of cluster centers.

[0016] Cluster center update formula:

[0017] Membership update formula:

[0018] In the formula: u ij For the first j The pixel belongs to the first i Membership degree of a class; m The weighted index, used to control the degree of ambiguity, is set to 2.0. kThis represents the upper limit of the number of iterations.

[0019] In step S2, by comparing the size of the reference object with the pixels in the image, a conversion relationship between the actual size and the pixels is established, and a scale bar is obtained based on the pixels and the actual size of the image.

[0020] In the formula: L For scale; n The number of pixels for reference; l The actual length of the reference object; By segmenting the sand and gravel aggregate particles in the image, and using the number of pixels and the length represented by a single pixel, combined with a scale bar, the feature size of the particles can be obtained. The specific method is as follows: The perimeter of the best-fit circumscribed ellipse of sand and gravel aggregate particles obtained using image recognition. C and geometric feature area A The major axis of the circumscribed ellipse that best fits the particle profile is obtained. a and short axis b Then, the equivalent particle size of the sand and gravel aggregate particles in the image is calculated. d :

[0021]

[0022] Calculate the equivalent particle size of sand and gravel aggregate. d for: .

[0023] The advantages and positive effects of this invention are as follows: It employs online detection, utilizes spatial coordinates to calculate the angle of each aggregate node from the origin, performs data normalization and perspective transformation to generate a highly robust sand and gravel aggregate feature map, and uses an eight-chain code algorithm to mark the target boundary and extract the particle size distribution of the sand and gravel aggregate. This allows for rapid and accurate online acquisition of the particle size of the sand and gravel aggregate, solving the problem of large differences in particle size and shape, which makes detection difficult. Online cleaning before detection solves the problem of impurities, overlap, noise, or artifacts in the captured images. The use of supplementary lighting equipment solves the problem of uneven lighting leading to changes in image brightness. Setting a reference object of known size in the shooting area solves the problem of inconsistent focal length or scale. In summary, this invention uses image recognition to perform morphological analysis of sand and gravel aggregate, enabling rapid and accurate online acquisition of the gradation characteristics of vibratory compaction stone pile sand and gravel aggregate, providing effective support for the rapid determination of sand and gravel aggregate gradation before vibratory compaction stone pile construction. Attached Figure Description

[0024] Figure 1 This is a flowchart of the present invention; Figure 2 This is an application scenario diagram of the present invention; Figure 3 This is a spatial distribution diagram of the sand and gravel aggregates in step S1.2 of the present invention; Figure 4 The aggregate gradation curve of the vibratory crushed stone pile detected by this invention is used for application.

[0025] In the picture: 1. Conveying equipment; 2. High-pressure spraying device; 3. Industrial camera. Detailed Implementation

[0026] To further understand the invention's content, features, and effects, the following embodiments are provided, along with detailed descriptions in conjunction with the accompanying drawings: Please see Figure 1 and Figure 2 A rapid detection method for sand and gravel aggregate gradation based on image recognition is proposed. A high-pressure spray device 2 is fixedly installed on the sand and gravel aggregate conveying equipment 1. An industrial camera 3 located downstream of the high-pressure spray device 2 is also fixedly installed. A reference object of known size and a supplementary lighting device (not shown in the figure) are also fixedly installed. The detection is then performed using the following steps: Step S1, Image morphology extraction, includes the following steps: Step S1.1, Aggregate washing During the construction of vibratory compaction stone piles, the sand and gravel aggregates are first laid on a conveyor belt and then transported to the pile hole along the conveying system. To improve the cleanliness of the aggregates, a high-pressure spraying device is installed at the end of the conveyor belt. Figure 2 As shown, water flow is used to wash the surface of the aggregate, removing adhering fine powder, mud, and other impurities. This washing process helps prevent the aggregate from clumping together due to mud adhesion, facilitating subsequent image recognition.

[0027] Step S1.2, Image Acquisition Ensuring the accuracy of the scale is crucial in calculating the particle size of sand and gravel aggregates. Therefore, when photographing sand and gravel aggregates, reference objects of known size must be included in the image to facilitate subsequent pixel-to-actual-size conversion. On the conveyor belt at the construction site, a 0.5-meter diameter wire ring is used as a reference object, while ensuring sufficient lighting during shooting to avoid shadows interfering with the clarity of the scale area. This not only improves the overall image quality but also significantly reduces the interference of external factors on the image recognition process.

[0028] Because high-definition cameras often cannot be positioned directly above the conveyor belt at construction sites, image distortion can occur, such as spatial misalignment or inconsistent scale. To minimize these errors and improve the accuracy of particle size distribution, multiple high-definition cameras are installed on the outer side above the conveyor belt. Image data captured from different angles is processed through spatiotemporal alignment to ensure a consistent time sequence across all perspectives, thereby guaranteeing the accuracy of aggregate particle size analysis.

[0029] Multiple high-definition cameras were used to capture images from different angles, and the resulting image data underwent spatiotemporal alignment processing to obtain a spatial distribution map of the aggregate. like Figure 3 The diagram shows the spatial distribution of aggregates of arbitrary shapes. Using a spatial coordinate system, the spatial location of the aggregates is expressed by the following formula: (1) In the formula: ( x , y , z () represents the three-dimensional spatial coordinates; n The number of nodes; O It is the origin of the three-dimensional space.

[0030] By calculating the lines connecting each node of the aggregate to the origin. S nO With the direction of the conveyor belt movement S OV The angle between F ( n , t We obtain the angle formula, which is expressed as follows: (2) Based on this, L ( n , t )and F ( n , t The data underwent normalization and perspective transformation to ensure numerical comparability and consistency of particle data from different samples. After angle adjustment and perspective transformation, the generated sand and gravel aggregate feature map exhibits higher robustness, providing strong support for subsequent image feature recognition and analysis.

[0031] Step S1.3: Image preprocessing In the rapid detection method of aggregate gradation for vibratory compaction stone piles based on image recognition, image preprocessing is a crucial step to ensure recognition accuracy and efficiency. Since the original image may be subject to various interferences during acquisition and transmission, common noise types such as Gaussian noise and impulse noise often affect image quality, thus impacting subsequent analysis and recognition results. Therefore, effective denoising is particularly important. Histogram equalization is a method in image processing, primarily used to improve image contrast, especially for images with low contrast and uneven grayscale distribution.

[0032] Each pixel in an image has a grayscale value (usually an integer between 0 and 255), and a histogram shows the frequency of each grayscale value. Histogram equalization adjusts the grayscale distribution of the image, making the pixel values ​​more evenly distributed across the entire grayscale range, thereby enhancing image contrast and detail, especially in both dark and bright areas. The `cv2.equalizeHist()` function maps pixel values ​​using a cumulative distribution function, the expression of which is: (3) In the formula: MN The total number of pixels in the image; CDF min This is the minimum value of the cumulative distribution; L This represents the number of gray levels.

[0033] By applying grayscale equalization preprocessing technology, the image can exhibit clearer and more accurate features in subsequent recognition tasks, which helps improve the clarity of areas that are difficult to distinguish. The optimization of image preprocessing provides more accurate data support for subsequent gradation detection, ensuring that the detection process of vibratory compacted stone pile aggregate can be carried out efficiently and smoothly.

[0034] Histogram equalization is performed using a function. Step S1.4: Image binarization based on K-means clustering The K-means algorithm is a distance-based clustering algorithm, also known as the K-means algorithm. This algorithm finds the optimal cluster centers by minimizing the variance of the clusters, and then classifies pixels in an image into two classes (target and background) based on these centers. For aggregate images, the target is typically sand and gravel aggregate, and the background is blank or other noisy areas.

[0035] Assume D={ x 1, x 2, …, x n Given a dataset of grayscale values ​​of all pixels in an image of sand and gravel aggregate to be clustered, the K-means algorithm will analyze each pixel... xj (1≤ j ≤ n ) assigned to C Categories, and through a membership matrix U ={ u ij} represents the membership relationship between each pixel and the cluster category, where u ij Indicates the first j The pixel belongs to the first i The degree of membership of each class.

[0036] The goal of the K-means algorithm is to find optimal cluster centers by minimizing the total squared error of clustering (i.e., the sum of the squared distances from each point to its cluster center). The calculation formula is as follows: (4) In the formula: v i For the first i Cluster centers of a class; | x j - v i |for the first j From the pixel to the i Euclidean clustering of cluster centers.

[0037] Cluster center update formula: (5) Membership update formula: (6) In the formula: u ij For the first j The pixel belongs to the first i Membership degree of a class; m The weighted index, used to control the degree of ambiguity, is set to 2.0. k This represents the upper limit of the number of iterations.

[0038] The K-means algorithm updates the membership matrix iteratively. U and cluster center v This process continues until the algorithm converges. During image binarization, the K-means algorithm divides image pixels into two main categories: target (sand and gravel aggregate) and background. Using the clustering results, cluster centers are directly used to separate the target from the background, and the grayscale value of the cluster centers is used as a threshold to achieve image binarization.

[0039] The formula used in the K-means algorithm is as follows: Step S1.5: Image boundary determination An octochain algorithm is used to find connected regions in the image, fill holes in the target, remove noise inside the target, and mark the target's boundaries.

[0040] Eight-chain coding is an encoding method used to represent image boundaries. It generates a series of direction values ​​based on the neighborhood of each boundary pixel. These direction values ​​are typically represented as changes along eight directions from the current pixel along the boundary.

[0041] An eight-chain code algorithm is used to process connected regions in binary images to identify and analyze the particle size distribution of sand and gravel aggregates. Specifically, boundary detection is performed by tracing the target region boundary pixel by pixel to generate a chain code sequence, ensuring that the connectivity of the target region is not incorrectly segmented. Holes in the target region are identified and filled to address potential voids in the particle distribution. More specifically, a connected component labeling algorithm is used to fill the holes, and area thresholding and morphological operations are combined to remove noise within the target region, further improving detection accuracy. The boundaries of the target are accurately extracted and visualized using visualization technology, which aids in subsequent gradation detection.

[0042] An octal chain code algorithm is used to process the binary image, extract the target boundary, identify and fill holes in the target area, remove noise inside the target area, and extract the particle size distribution of sand and gravel aggregates. Step S2, Grain Distribution Detection The acquired images show that most of the sand and gravel aggregates on the conveyor belt exhibit a near-elliptical shape. To accurately reflect the true three-dimensional morphology of the sand and gravel aggregates, the equivalent diameter of the ellipsoidal aggregate particles was used. d The gradation curve of the sand and gravel aggregate is calculated based on the particle size range. The specific method is as follows: By comparing the size of a reference object with the pixels in the image, a conversion relationship between actual size and pixels is established, thus ensuring accurate conversion. The scale is obtained based on the pixels and the actual size of the image. (7) In the formula: L For scale; n The number of pixels for reference; l The actual length of the reference object.

[0043] By segmenting the sand and gravel aggregate particles in the image, and using the number of pixels and the length represented by a single pixel, combined with a scale bar, the feature size of the particles can be obtained. The specific method is as follows: The perimeter of the best-fit circumscribed ellipse of sand and gravel aggregate particles obtained using image recognition. C and geometric feature area A The major axis of the circumscribed ellipse that best fits the particle profile is obtained. aand short axis b Then, the equivalent particle size of the sand and gravel aggregate particles in the image is calculated. d : (8) (9) Calculate the equivalent particle size of sand and gravel aggregate. d for: (10) Count the number of particles within each particle size range, and calculate the volume of each particle using the volume formula for an ellipsoid.

[0044] Assuming all sand and gravel particles have the same density, convert volume to mass. Multiply the number of particles within each particle size range by the mass of a single particle to obtain the total mass of that particle group. Based on the percentage of each particle group's mass in the total mass, plot the gradation curve of the vibratory compacted stone pile aggregate.

[0045] This invention uses image recognition to perform morphological analysis on sand and gravel aggregates, ensuring that the three-dimensional characteristics of the particles can be accurately reflected, thereby providing support for the rapid detection of sand and gravel aggregate gradation in vibratory compaction stone piles.

[0046] Although preferred embodiments of the present invention have been described above in conjunction with the accompanying drawings, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other modifications under the guidance of the present invention without departing from the spirit and scope of the claims, and all of these modifications are within the scope of protection of the present invention.

Claims

1. A rapid detection method for sand and gravel aggregate gradation based on image recognition, characterized in that, A high-pressure spray device, an industrial camera located downstream of the high-pressure spray device, a reference object of known size, and supplementary lighting equipment are fixedly installed on the sand and gravel aggregate conveying equipment. Then, the following steps are used for inspection: Step S1, Image Morphology Extraction, includes the following sub-steps: Step S1.1: Aggregate cleaning. Turn on the high-pressure spray device to rinse the surface of the aggregate with water. Step S1.2: Image acquisition. An industrial camera captures images of sand and gravel aggregates. The images contain reference objects to determine the scale. Using spatial coordinates, the angles of each aggregate node to the origin are calculated. Data normalization and perspective transformation are performed to generate a highly robust sand and gravel aggregate feature map. Step S1.3: Image preprocessing, first remove noise, and then use histogram equalization to preprocess the image. Step S1.4: Image binarization. The K-means algorithm is applied to cluster the sand and gravel aggregate image, dividing the pixels into two categories: target and background. The target is the sand and gravel aggregate. The gray value of the cluster center is used for image binarization. Step S1.5: Image boundary determination. The eight-chain code algorithm is used to find connected regions in the image, fill the holes of the target, remove noise inside the target, mark the boundary of the target, and extract the particle size distribution of sand and gravel aggregate. Step S2, gradation detection: The aggregate morphology is analyzed by image recognition, the equivalent diameter of the fitted ellipsoidal aggregate particles is calculated, and the volume and mass of the particles are calculated by combining the scale and geometric features. Finally, the gradation curve of the vibratory compaction stone pile sand and gravel aggregate is drawn.

2. The rapid detection method for sand and gravel aggregate gradation based on image recognition according to claim 1, characterized in that, The industrial camera includes multiple industrial cameras with different shooting angles. In step S1.2, the multiple industrial cameras take pictures of sand and gravel aggregates from different angles, perform spatiotemporal alignment processing on the obtained image data, and then obtain a spatial distribution map of the aggregates.

3. The rapid detection method for sand and gravel aggregate gradation based on image recognition according to claim 1, characterized in that, In step S1.2, the spatial distribution map of the aggregate is obtained using an image, and the spatial location of the aggregate is expressed using the following formula based on a spatial coordinate system: In the formula: (x,y,z) represents the three-dimensional spatial coordinates; n represents the number of nodes; O represents the origin of the three-dimensional space; By calculating the lines connecting each node of the aggregate to the origin. S nO With the direction of the conveyor belt movement S OV The angle between F ( n , t We obtain the angle formula, which is expressed as follows: The L(n,t) and F(n,t) data are normalized and transformed by perspective. After angle adjustment and perspective transformation, a highly robust sand and gravel aggregate feature map is generated.

4. The rapid detection method for sand and gravel aggregate gradation based on image recognition according to claim 1, characterized in that, In step S1.3, the histogram equalization process uses the cv2.equalizeHist() function to map pixel values ​​using a cumulative distribution function, the expression of which is: In the formula: MN is the total number of pixels in the image; CDFmin is the minimum value of the cumulative distribution; L is the number of gray levels.

5. The rapid detection method for sand and gravel aggregate gradation based on image recognition according to claim 1, characterized in that, The calculation formula for the K-means algorithm in step S1.4 is as follows: In the formula: v i For the first i Cluster centers of a class; | x j - v i |for the first j From the pixel to the i Euclidean clustering of cluster centers. Cluster center update formula: Membership update formula: In the formula: u ij For the first j The pixel belongs to the first i Membership degree of a class; m The weighted index, used to control the degree of ambiguity, is set to 2.

0. k This represents the upper limit of the number of iterations.

6. The rapid detection method for sand and gravel aggregate gradation based on image recognition according to claim 1, characterized in that, Step S2 establishes a conversion relationship between actual size and pixels by comparing the size of the reference object and the pixels in the image, and obtains a scale bar based on the pixels and the actual size of the image. In the formula: L For scale; n The number of pixels for reference; l The actual length of the reference object; By segmenting the sand and gravel aggregate particles in an image, and using the number of pixels and the length represented by a single pixel, combined with a scale bar, the characteristic size of the particles can be obtained. The specific method is as follows: The perimeter of the best-fit circumscribed ellipse of sand and gravel aggregate particles obtained using image recognition. C and geometric feature area A The major axis of the circumscribed ellipse that best fits the particle profile is obtained. a and short axis b Then, the equivalent particle size of the sand and gravel aggregate particles in the image is calculated. d : Calculate the equivalent particle size of sand and gravel aggregate. d for: 。

Images